A convolutional neural network model for prediction of pressure-drop at wellhead in geothermal power plant

The objective of the research is to develop an optimal management and control framework to contribute to improving the capacity factor of the geothermal power plant. As the initial consideration of the study, the article aims to identify the phenomenon that causes to decrease capacity factor from the analysis of a process data, to detect the phenomenon to support the decision-making of operators. For achieving the latter goal, we construct the framework that announces the operator to how much anomaly with a convolutional neural network model, which can capture non-stationary and nonlinearity of process data. Through the literature survey and the analysis of the process data, we confirmed that the steam flow rate and pressure of wellbore having two feed zones oscillate, and the wellbore reached the spurts stop of the steam due to oscillation growth. We confirmed that the pressure-drop events occurred six times in 2018, and it would be one of the causes to decrease capacity factor. As numerical results of the proposed framework, before the pressure-drop happens, they are increased that the amplitude of the second harmonic wave, the period of the first harmonic wave, and the anomaly score we derived. We could conclude that the pressure-drop, as mentioned above indexes made by the proposed method, could predict the pressure-drop at the one minute ahead, even in the case that pressure time-series have the nature of non-stationarity and nonlinearity.